|
Alereon posted:AMD is actually working pretty closely with ARM. While I don't think they'll make ARM CPUs, there's been talk that they'll produce a licensable GPU block for ARM CPUs, much like PowerVR. Such as Imageon, which was sold to Qualcomm and became Adreno?
|
# ? Nov 6, 2011 00:54 |
|
|
# ? Dec 6, 2024 16:09 |
|
HalloKitty posted:Such as Imageon, which was sold to Qualcomm and became Adreno?
|
# ? Nov 6, 2011 05:32 |
|
Bulldozer Opterons were launched today, we're waiting for reviews to get written. Basically, AMD will deliver a 16-core Interlagos Opteron at the same clockspeed and TDP that you would have gotten a 12-core Magny-Cours Opteron at. A guy on the Real World Tech forums ran some numbers, the new Opterons are about 7-15% slower per-core at the same TDP, but 5-29% faster on a per-socket basis at the same TDP. A welcome improvement compared to their old Opterons, but not really enough to start to catch with Xeons, especially since Xeons already had the per-core performance crown.
|
# ? Nov 15, 2011 10:18 |
|
Is there any news on the 28nm gpu lineup? Southern Islands is the chip family codename right? I haven't heard anything on them in months and thought they were supposed to be out by now? Granted it feels like the next round of nvidia boards are in the same boat.
|
# ? Nov 15, 2011 19:44 |
|
Fuzzy Mammal posted:Is there any news on the 28nm gpu lineup? Southern Islands is the chip family codename right? I haven't heard anything on them in months and thought they were supposed to be out by now? Granted it feels like the next round of nvidia boards are in the same boat. I bet that if anything's holding them up it's yields on the new 28nm process.
|
# ? Nov 15, 2011 20:17 |
|
Anandtech's Bulldozer for Servers: Testing AMD's "Interlagos" Opteron 6200 Series review is out.Fuzzy Mammal posted:Is there any news on the 28nm gpu lineup? Southern Islands is the chip family codename right? I haven't heard anything on them in months and thought they were supposed to be out by now? Granted it feels like the next round of nvidia boards are in the same boat.
|
# ? Nov 16, 2011 04:34 |
|
If I had to guess, we'll see Southern Islands right before CES.
|
# ? Nov 16, 2011 04:37 |
|
What an article title from Ars: AMD's Bulldozer server benchmarks are here, and they're a catastrophe
|
# ? Nov 22, 2011 04:28 |
|
I'm no AMD fanboy (OP of the Sandy Bridge thread checking in), but that analysis seems really bad. Frankly, Ars Technica is a lifestyle blog, they gave up tech years ago and this article shows they don't have the expertise to draw meaningful conclusions. They're analyzing Bulldozer in terms of per-core, per-clock performance, when it needs to be analyzed in terms of per-socket performance. They're looking at benchmarks from bizarrely configured multi-million dollar server installations, when as Anandtech points out the actual real servers that people would buy give AMD a HUGE price advantage, especially when large amounts of RAM are involved. They also called out Bulldozer for poor power efficiency in virtualized applications, when Anandtech showed that this was an ESX issue. It's not even anything complicated like thread scheduling, just that enhanced C-states are disabled by default. This is something that's pretty easily fixable, either yourself or through an ESX update. Really, Bulldozer for servers isn't the hit out of the park that people were hoping for, but it IS a tangible incremental improvement over Magny-Cours at the applications that people buy AMD processors for, which are virtualization farms and other situations where you want to cram as many cores and/or RAM into a given power/money/space budget as possible, and don't really care about per-thread performance. It doesn't really change the landscape much, except for the addition of encryption/decryption acceleration.
|
# ? Nov 22, 2011 07:31 |
|
Maybe I'm not an excitable tech site writer looking for page hits, but 'marginally quicker in some things, a bit slower in others and a bit more power hungry for a given price point' isn't what I'd call a 'catastrophe'.
|
# ? Nov 22, 2011 09:35 |
|
Mr Chips posted:Maybe I'm not an excitable tech site writer looking for page hits, but 'marginally quicker in some things, a bit slower in others and a bit more power hungry for a given price point' isn't what I'd call a 'catastrophe'. I think you forgot the "3+ years wait". That makes it one.
|
# ? Nov 22, 2011 10:24 |
|
freeforumuser posted:I think you forgot the "3+ years wait". That makes it one. More like "today's forecast Intel's foot remains on AMD's throat in the server market" makes it one. If AMD were the only processor manufacturer this would be a disappointing and confusing release, but they're in a starkly competitive environment and they are going to get screwed if they can't put a hell of a lot of these new boondoggles in computers smart buyers probably aren't going to want to stick 'em in.
|
# ? Nov 22, 2011 11:46 |
|
AMD has canceled their Wichita and Krishna APUs, the successors to Brazos, because Edit: It actually was GloFo and not TSMC, my bad! Alereon fucked around with this message at 12:47 on Nov 23, 2011 |
# ? Nov 22, 2011 11:52 |
|
Alereon posted:AMD has canceled their Wichita and Krishna APUs, the successors to Brazos, because TSMC is unable to get the 28nm process working within a reasonable timeframe. Instead, they'll refresh Brazos on 40nm with slightly improved clockspeeds and turbo. They're renaming the integrated graphics to the Radeon HD 7300-series, but it's still VLIW5 so I doubt there's any meaningful changes. The real changes are in the new chipset, which will include SATA600 (2 ports) and USB 3.0 (2 ports). There is always more, and it is always worse, jeeeeesus christ Is AMD involved in the downfall of humanity in some way we just can't see right now? Is there a time traveler doing everything possible to ensure that AMD fails?
|
# ? Nov 22, 2011 12:27 |
|
They're just a bystander. Skynet was built on TSMC's 28nm process.
|
# ? Nov 22, 2011 12:30 |
|
How much of this is a result of AMD spinning off their fabs?
Mr Chips fucked around with this message at 12:36 on Nov 22, 2011 |
# ? Nov 22, 2011 12:31 |
|
On the plus side (well, minus side for us) Intel's Cedar Trail Atoms are complete poo poo, they have an improved on-die GPU with DirectX 10.1 support, but Intel decided to give up on trying to write drivers. They're releasing DirectX 9 only drivers for Windows 7 32-bit only, there will be no 64-bit support or exposure of the DX10.1 featureset until at least Windows 8. It'll be 2013 and the Silvermont architecture before Intel has anything to answer Brazos with.
|
# ? Nov 22, 2011 12:33 |
|
Mr Chips posted:How much of this is a result of AMD spinning off their fabs? 28nm has been hard for everyone, and they were having trouble getting 32nm going for Llano so AMD should have saw this coming. That fact gives me pause about the idea that they would go with them for Wichita and Krishna, especially since TSMC built Brazos. It just seems weird to switch like that generation wise. edit: completing a thought. Ryokurin fucked around with this message at 15:25 on Nov 22, 2011 |
# ? Nov 22, 2011 14:29 |
|
Y'know, one thing that's been bugging my uninformed-about-chip-fab self about Bulldozer has been AMD's recommended overclocking voltages. In [H]'s Bulldozer review, it quoted AMD as saying to overclock by setting a Vcore from 1.4 to 1.55 V. It took a little re-evaluation, but Intel recommended a top voltage of 1.38 V for their 32nm processors. Is AMD's process different, or are overclocked Bulldozer chips going to burn out faster than a fart in a frat house?
|
# ? Nov 22, 2011 14:46 |
|
Wow, these new desktop FX chips are garbage eh? Sad, the $130 FX-4100 had some potential to be a serious overclocker (apparently very easy to get to 4.6GHz with no aftermarket cooling, and 5GHz is relatively easy to attain with some basic gear) but even then they'd still get stomped...
|
# ? Nov 22, 2011 15:47 |
|
Factory Factory posted:Is AMD's process different, or are overclocked Bulldozer chips going to burn out faster than a fart in a frat house? Hard to find good public information but turned this up. Paul DeMone posted:AMD slide presentation Not exactly confidence inspiring. PC LOAD LETTER fucked around with this message at 16:05 on Nov 22, 2011 |
# ? Nov 22, 2011 15:56 |
|
VERTiG0 posted:Wow, these new desktop FX chips are garbage eh? Sad, the $130 FX-4100 had some potential to be a serious overclocker (apparently very easy to get to 4.6GHz with no aftermarket cooling, and 5GHz is relatively easy to attain with some basic gear) but even then they'd still get stomped... There are a couple tasks where they are close to the i7 but for 90% of your every day tasks they aren't much faster (or even slower) than the X6
|
# ? Nov 22, 2011 15:58 |
|
VERTiG0 posted:Wow, these new desktop FX chips are garbage eh? Sad, the $130 FX-4100 had some potential to be a serious overclocker (apparently very easy to get to 4.6GHz with no aftermarket cooling, and 5GHz is relatively easy to attain with some basic gear) but even then they'd still get stomped... Power consumption makes the question of how good they are at overclocking more complex. For starters, the FX 8150 BE (3.6GHz stock) draws about 240W stock, compared to 130ish for the 2600K and 2500K at their 3.4GHz and 3.3GHz stock speeds. The FX 8150 goes over 400W once you get them to 4.5GHz+, compared to Sandy Bridge which are hanging out around 200W at 4.6GHz. That's when really loaded down, obviously, but still, the power consumption comparison and clock for clock performance deficit means why would anyone build an FX chip system if they aren't an AMD supporter regardless?
|
# ? Nov 22, 2011 16:07 |
|
Mr Chips posted:Maybe I'm not an excitable tech site writer looking for page hits, but 'marginally quicker in some things, a bit slower in others and a bit more power hungry for a given price point' isn't what I'd call a 'catastrophe'. Considering that Interlagos is more power hungry, likely more expensive to make (due to the massive die size of each Orochi die), and has the advantage of being a process node ahead of Magny Cours, the fact that it's only sometimes faster is really damning. Contrast this with Intel's ability to get ~20% performance increases from generation to generation without a process shrink and that makes AMD's failure even more stark. If having a product that's late, slow, hot and costly to make while your competition is moving forward almost without fault isn't a catastrophic design failure then I'm not sure what is.
|
# ? Nov 22, 2011 17:16 |
|
TMSC having process issues? Well I never!
|
# ? Nov 22, 2011 19:06 |
|
Factory Factory posted:Is AMD's process different, or are overclocked Bulldozer chips going to burn out faster than a fart in a frat house? Not that the two conditions are necessarily exclusive, but yes, AMD's process (and many other potentially relevant aspects of the overall design) is significantly different from Intel's.
|
# ? Nov 22, 2011 19:47 |
|
rscott posted:TMSC having process issues? Well I never! Looks like it's the opposite. They are droping Global for TSMC. http://www.extremetech.com/computing/106217-manufacturing-bombshell-amd-cancels-28nm-apus-starts-from-scratch-at-tsmc
|
# ? Nov 22, 2011 23:22 |
|
VERTiG0 posted:Wow, these new desktop FX chips are garbage eh? Sad, the $130 FX-4100 had some potential to be a serious overclocker (apparently very easy to get to 4.6GHz with no aftermarket cooling, and 5GHz is relatively easy to attain with some basic gear) but even then they'd still get stomped... Forget anything from Intel or even the PhII X6s, that thing has trouble beating a ~3 year old X4 955BE, both stock or when overclocked. It makes even the lowly $130 price tag look overpriced. http://www.legionhardware.com/images/review/AMD_FX-8150_FX-8120_FX-6100_and_FX-4170/Overclocking_02.png freeforumuser fucked around with this message at 00:33 on Nov 23, 2011 |
# ? Nov 23, 2011 00:30 |
|
freeforumuser posted:Forget anything from Intel or even the PhII X6s, that thing has trouble beating a ~3 year old X4 955BE, both stock or when overclocked. It makes even the lowly $130 price tag look overpriced. The marketing team that decided to pretend their modules were cores really, really, really screwed up. It makes comparisons look godawful. They should have gone with conventional and comparable marketing to other current and in-production hardware. Nothing to be ashamed of for selling a 4-core processor with nifty hardware-based dual threading per core. They could have really capitalized on how it's not like Hyperthreading where it's barely hardware at all, just presenting 4 real and 4 logical, no, this is real, hardware dual-thread per core! But instead they went with the deeply misleading modules/cores approach, and so people look at what AMD are calling 4-core processors and expect them to perform competitively with other 4-core processors. Then they don't, because they're actually 2-core processors with a nifty hardware element to their dual threading but it's not the same thing as full blown cores... Which makes them look really awful by comparison. Even the ad speech about AMD bringing the first 8-core desktop processors to market is baffling and stupid since performance comparisons show the only thing remotely 8-core about the higher end FX-line is their power consumption, which does start to look like octo-core setups. Horrible idea, way oversold the concept, could have been done with so much more finesse. Aaarrgh AMD you're killing me here.
|
# ? Nov 23, 2011 01:46 |
|
It's perfectly legitimate to call a 4-module Bulldozer processor an 8-core processor. There's nothing fundamentally wrong with this approach, it's just that Bulldozer had so much cache that it ran up against the limits of how big you can make a CPU, thus preventing AMD's designers from spending the transistors necessary to actually perform well in the real world. Anandtech's Bulldozer article goes into some detail about the compromises AMD made both for die size and in the pursuit of clock speed, unfortunately the latter couldn't pay off at all because Global Foundries was unable to deliver to expectations on their 32nm process. On the plus side, this does mean that when AMD refreshes Bulldozer for the desktop without the L3 cache (and presumably on a more mature 32nm process), it could see meaningful improvements.
|
# ? Nov 23, 2011 02:16 |
|
Don't worry I know 5 people that are gamers with no clue how computers work. The important factors to them in order are Number of Cores VRAM RAM HDD Space I was able to talk 1 out of buying a bulldozer in the upcoming month or when computers will "be on sale" they have no idea how to build, I guess the other 4 I don't know very well so they don't trust my judgement or something. If I was still playing WoW I'm sure the number would be higher because everyone over 30 that plays seems to fall into the I want a new computer each year that has more cores category. I think AMD will do okay because of the the current more cores is better hype going on with the mainstream. gently caress my dad asked me when 5, 6 and 7 core processors came out because he heard about a new 8 core and wanted it. (He only browses the internet and is using under 10% of everything on that system, its old parts I just threw together to stop him from buying something from a store).
|
# ? Nov 23, 2011 02:22 |
|
pixaal posted:Don't worry I know 5 people that are gamers with no clue how computers work. The important factors to them in order are I doubt AMD can even sell enough BDs to the unsuspecting retail consumers or AMD fanboys when the FX-8150 is still OOS at Newegg, let alone finding itself into mass-market Dells/HPs. freeforumuser fucked around with this message at 03:16 on Nov 23, 2011 |
# ? Nov 23, 2011 03:13 |
|
They've landed a number of supercomputer deals for Bulldozer-based Opterons, each of those is a not-insignificant number of dies (thousands of processors, two dies per processor) sold at substantially above consumer price levels. The only thing you're going to see branded as an FX-series processor are either defective dies that couldn't be sold as Opterons or excess supply above and beyond what's needed for their server business. Since yields are still very low it makes sense there wouldn't be much available for consumers.
|
# ? Nov 23, 2011 03:18 |
|
I was looking at the BD cores from a commercial perspective recently. I could easily see the cores working well for niche supercomputer applications or specifically optimised code. I gather that market is priced based on actual performance rather than marketing.
|
# ? Nov 23, 2011 03:31 |
|
Alereon posted:It's perfectly legitimate to call a 4-module Bulldozer processor an 8-core processor. I wouldn't say it's 100% legitimate. I'd be happier if they called it 4+4 cores or something weird like that. Alereon posted:They've landed a number of supercomputer deals for Bulldozer-based Opterons, each of those is a not-insignificant number of dies (thousands of processors, two dies per processor) sold at substantially above consumer price levels. But does it make any business sense? I can't imagine any deal that would cover the R&D costs of developing a chip like this.
|
# ? Nov 25, 2011 00:04 |
|
How about deals with Cray? ..the XK6 is capable of scaling to 500,000 Opteron cores...
|
# ? Nov 25, 2011 08:41 |
|
pixaal posted:...fall into the I want a new computer each year that has more cores category. This is not a bad thing IMO. These people will spread their wacky logic to others in their circle of friends and family and hopefully game developers won't be so shy to actually leverage their engines against more cores. Its a catch-22 which needs these people to push things forward.
|
# ? Nov 25, 2011 15:56 |
|
Shaocaholica posted:This is not a bad thing IMO. These people will spread their wacky logic to others in their circle of friends and family and hopefully game developers won't be so shy to actually leverage their engines against more cores. Its a catch-22 which needs these people to push things forward. Thing is, quad-core CPUs are by no means universal: link. About as many Steam users are still using dual-core CPUs as are using quad-cores. Folks with 8-core BD CPUs will be long, long, long in waiting for their market segment to have enough representation for 4 vs. 8 threads to be worth a programmer's time.
|
# ? Nov 25, 2011 16:13 |
|
Factory Factory posted:Thing is, quad-core CPUs are by no means universal: link. About as many Steam users are still using dual-core CPUs as are using quad-cores. Folks with 8-core BD CPUs will be long, long, long in waiting for their market segment to have enough representation for 4 vs. 8 threads to be worth a programmer's time. Its still progress IMO. You'll always have these guys leading the pack. If having one 8+ core zealot means 3 dual core buddies upgrade to quad or above, then I think they are doing everyone a service. Again, its a catch 22. No developer is going to optimize for quad core if no one has them and not many gamers are going to upgrade unless they can get something out of it. Shaocaholica fucked around with this message at 16:34 on Nov 25, 2011 |
# ? Nov 25, 2011 16:31 |
|
|
# ? Dec 6, 2024 16:09 |
|
Mr Chips posted:Maybe I'm not an excitable tech site writer looking for page hits, but 'marginally quicker in some things, a bit slower in others and a bit more power hungry for a given price point' isn't what I'd call a 'catastrophe'.
|
# ? Nov 25, 2011 22:26 |